BKS News
English / Deutsch
AI-generated illustration: Cloud Computing and AI Infrastructure
Image generated with Pollinations.ai
Weekly Briefing 7 min read

AI Weekly - Week 49/2025

Sunday, December 7, 2025

ai weekly anthropic aws openai bun claude-code reinvent trainium agents
This article was researched and written with AI

TL;DR

This week in 30 seconds:

  • Anthropic Milestone: First M&A deal (Bun) + Claude Code reaches $1B ARR in just 6 months
  • AWS re:Invent: Frontier Agents (Kiro), Trainium3 Chips, and 18 new Bedrock models
  • Code Red: Sam Altman sends panic memo to OpenAI employees over Gemini competition

The Story of the Week

Anthropic Makes Its First Deal – And It’s Strategically Brilliant

On December 2, 2025, something historic happened: Anthropic made its very first acquisition. [1] The target? Bun – the JavaScript runtime that started as a faster alternative to Node.js and now combines a package manager, bundler, and test runner in a single executable. [4]

The price wasn’t officially disclosed, but The Information reports a deal “in the low hundreds of millions.” [1] That sounds like a lot – until you understand why Anthropic is doing this.

Claude Code, Anthropic’s AI coding tool, ships as a Bun executable to millions of users. [4] The runtime isn’t an optional component; it’s the foundation. If Bun breaks, Claude Code breaks. This dependency on an external open-source project was a strategic risk that Anthropic has now eliminated.

The timing is no coincidence: On the same day, Anthropic announced that Claude Code reached $1 billion annual run rate – just six months after its public launch in May 2025. [1] Netflix, Spotify, KPMG, L’Oreal, and Salesforce are already using it. [1]

“Bun will remain open source and MIT-licensed, and we will continue to invest in making it the runtime, bundler, package manager, and test runner of choice for JavaScript and TypeScript developers.” - Anthropic Blog [1]

Critical Voices: Not everyone is thrilled. The Zig community (Bun is built on Zig) has strict no-LLM policies in its code of conduct. And developers are asking: Will Bun now be optimized for Claude Code instead of the broader JS community? [4]

Bun founder Jarred Sumner and his team remain on board and continue working on the project. [4] The difference: Now with Anthropic’s resources backing them.

Bottom Line: When your $1B product runs on open-source infrastructure, you buy the team. Anthropic secures control over its own toolchain – and shows that the AI coding wars are now being fought at the infrastructure level.

More Top Stories

AWS re:Invent 2025: The Era of Frontier Agents Begins

From December 1-5, AWS transformed Las Vegas into an AI hub once again. [2] The clear message: Agents are the new paradigm.

Three “Frontier Agents” stole the show: [2]

  • Kiro Autonomous Agent: A virtual developer that maintains context across sessions and works asynchronously. The trick: It never forgets and learns from your feedback.
  • AWS Security Agent: Your personal security advisor that analyzes threats and suggests fixes.
  • AWS DevOps Agent: 24/7 ops support without a human on-call team.

On the hardware side: Trainium3 on 3nm technology delivers 4.4x more compute performance with 4x better energy efficiency. [2] And a Trainium4 preview promises another doubling. Fun fact: Anthropic is already using 500,000 Trainium2 chips and plans to double that by year-end under Project Rainier. [2]

Amazon Nova 2 brings three new models: Pro (Reasoning), Sonic (Voice), Omni (Multimodal). Plus 18 new open-weight models in Bedrock – the largest model expansion ever. [2]

OpenAI Declares “Code Red”

On December 2, Sam Altman sent an internal memo with the subject “Code Red.” [3] The message: ChatGPT is losing ground to Gemini.

The numbers behind it sound alarming: Gemini is growing faster than ChatGPT (measured by MAU since August). [3] Google reached 650 million monthly users, and Anthropic is gaining ground in the enterprise market – over 300,000 business customers, up from fewer than 1,000 two years ago. [3]

But context matters: OpenAI calls “Code Red” more often than you’d think. Mark Chen confirmed: “We do this more frequently.” [3] And Altman himself expects to have solved the problem by January. ChatGPT still has 800 million weekly users – absolute dominance remains. [3]

OpenAI’s response is nonetheless serious: Advertising plans postponed, Health/Shopping/Advertising features delayed. [3]

The irony: Google experienced exactly this “Code Red” situation in November 2022 when ChatGPT launched. Sundar Pichai admitted internally that Google hadn’t “quite gotten to a level where you could put it out.” [3] Three years later, the roles have reversed.

Quick Hits

Briefly noted:

  • Snowflake-Anthropic $200M Deal: Partnership for agentic AI in enterprises – Claude goes deep into the Data Cloud. [5] Details
  • Claude for Nonprofits: Free AI training and discounted Claude access for nonprofit organizations. Program
  • Lyft AI Agent: 87% faster problem-solving through Claude via Amazon Bedrock – and 70% more driver engagement. [2] Case Study
  • Bedrock AgentCore Memory: Agents can now learn and store user preferences over time. [2] AWS Blog
  • Nova Act GA: Amazon’s browser automation agent now generally available – 90%+ reliability for enterprise deployments. [2] More

Tool of the Week

Kiro Autonomous Agent - The Developer Who Never Sleeps

AWS introduced Kiro as a preview – an autonomous agent that fundamentally differs from previous AI assistants. [2] The key differentiator: Persistent Context.

While ChatGPT or Claude restart with each session, Kiro maintains awareness across all your work. It remembers decisions, learns from your feedback, and can execute tasks asynchronously in the background while you work on something else. [2]

Who should use it: Teams that want to automate recurring development tasks – from code reviews to dependency updates to documentation. AWS Kiro

Fail of the Week

MIT Study: 95% of Enterprise AI Pilots Deliver Zero Value

While AWS showcases impressive demos and Anthropic celebrates milestones, a sobering reality emerges from research: A 2025 MIT study shows that 95% of companies that launched AI pilots see no measurable business value. [6]

Even harsher: The METR study from July 2025 found that experienced developers working with AI tools are 19% slower than without them. [7] The reason? “Fast right” code that costs more debug time than it saves.

Root Cause: The gap between demo and production is massive. Tools work in controlled environments but fail at enterprise complexity.

What We Learn: Not every glossy launch means real business impact. The question isn’t “What can AI do?” but “What does AI deliver in your specific context?”

Number of the Week

$1 Billion ARR in 6 Months

Claude Code’s growth rate – from zero to $1B annual run rate between May and December 2025 [1]

For comparison: Slack took 5 years to reach this milestone. Zoom took 9 years. Anthropic overall expects $9B ARR by end of 2025 and projects $20-26B for 2026. [1] The AI coding market isn’t just hot – it’s exploding.

Reading List

For the weekend:

  1. Anthropic acquires Bun - The official announcement with integration details (3 min)
  2. AWS re:Invent 2025 Recap - All major announcements at a glance (8 min)
  3. OpenAI Code Red Story - The full story behind Altman’s panic memo (6 min)

Next Week

What’s coming:

  • MCP Announcement expected – rumors about open-source donation
  • Anthropic-Accenture Deal being finalized for enterprise AI production
  • OpenAI Response to Code Red – new features or model updates?

This newsletter was researched and written AI-assisted. Hero image generated with Pollinations.ai.